Antibodies and antigens: It's all about the numbers game

Antibodies are exquisitely specific molecules that recognize and eliminate harmful agents bearing foreign pathogenic and disease antigens. Today, close to 100 therapeutic antibodies are in clinical trials and hundreds more in preclinical development. They are being used for treating a range of illnesses including inflammation, autoimmune diseases, cancer, cardiovascular diseases, allergic disorders, and infectious diseases. The focus of the antibody field has now essentially shifted from discovery to product development and, indeed, this area represents a triumph for the biotechnology industry. However, the selection of antibodies can be a laborious process, because this is normally done one antigen at a time. The article by Bowley et al. (1) in a recent issue of PNAS is pioneering and demonstrates how to accelerate the identification of antibodies to a multitude of antigens.

It is now clear that the discovery of combinatorial antibody libraries has revolutionized immunochemistry (2–8). Of profound significance is that there is no longer any need to use immunization procedures to produce antibodies—some 100 or so years after antibodies were discovered in 1890! Importantly, antibody libraries also allow the construction of immunological repertoires that are at least comparable in size with those of Nature. Moreover, libraries such as phage, yeast, Escherichia coli surface, etc., unlike their natural counterparts, are not restricted by the constraints of self-tolerance. This is especially important because most of the therapeutic antibodies in the clinic are antibodies to self. Without a doubt, antibody libraries have profound implications for human health. The first dramatic example is Humira (Human monoclonal antibody in rheumatoid arthritis), which is an antibody generated by using phage display technology, and used by thousands of patients worldwide with rheumatoid arthritis. It works by binding, with high affinity, to its antigen, tumor necrosis factor alpha (TNF-α), thus preventing it from activating TNF-α receptors that are important in inflammatory reactions.

The ability to identify specific antibody–antigen pairs rapidly is therefore of enormous significance. For instance, as Bowley et al. (1) point out, the human and other genome projects provide opportunities to generate high-affinity monoclonal antibodies to every protein in the genome. The problem may actually be even larger than the authors suggest, because each protein in the genome (of ≈30,000 proteins) will be characterized by many epitopes (typically each including a handful of amino acids). What will then be required is the simultaneous selection of monoclonal antibodies to a large set of antigens, rather than the current approach of selecting one antigen at a time. The essence of the approach in the article by Bowley et al. is to mix an antibody library with an antigen library and retrieve specific antibody–antigen pairs (Fig. 1).

Fig. 1.

Strategy for combinatorial selection of replicating antibody–antigen pairs. The two libraries (displayed on phage or yeast, for example) are mixed, and specific antibody–antigen pairs are fluorescently labeled and sorted by using flow cytometry. Separation and selection allows further enrichment.

The authors use two different display platforms for the antibody and antigen libraries. The requirement of each platform is stringent in that they must be able replicate independently. In this first example the two platforms were phage and yeast. Many variations are possible; for instance, it would be possible to humanize the glycosylation in yeast so as to allow glycoproteins, which constitute the majority of extracellular targets for therapeutic antibodies. The use of a yeast platform enables expression of domains or whole proteins or even protein fragments.

To demonstrate the proof of concept, the authors expressed on the surface of yeast a collection of single-chain Fv molecules (which comprise the antibody binding sites) from an HIV-infected individual, as the antibody library. The antigen library, expressed on phage, contained peptide fragments of the HIV-1 gp160 protein. It was known previously that a single-chain Fv, termed Z13, recognized a linear epitope on the HIV-1 gp160 protein that was included in a 36-aa peptide termed TJ1D. To optimize the conditions for cognate antigen–antibody selection, the libraries were “spiked” with both Z13 in the antibody library and with TJ1D in the antigen phage library at a frequency of 1:10⁴, making the frequency of the cognate pair 1:10⁸.

The problem of detection of the complex means that several rounds of selection may be required (Fig. 1), but the enrichment resulting from this is impressive. (There is a beautiful control showing the specificity of the cognate pairs. When the TJ1D peptide has the D changed to N to give the mutated peptide TJ1N, there is no enrichment.) There are other technical problems that are discussed, but the format and ideas are clear.

The experiments reported here expressed the antibody library on yeast and the antigen one on phage. As Bowley et al. (1) point out it is probably better to reverse this, as the size of the yeast library is limited compared with phage. The size of the combinatorial libraries in phage should also allow detection of many different antibodies to each target antigen. In addition to the generation of therapeutic antibodies the high-throughput screen described in this paper will also allow the detection of cell surface markers and diagnostics, which will be of immense value to the research community. Although it is, in principle, possible to generate antibodies to every part of the genome, one should probably start with a collections of antigens likely to impact human therapy. These could include lymphokines, cytokines, G protein-coupled receptors, as well as immunologically difficult targets, such as the self-like epitopes found on the surface of tumors and some pathogens.